Skip to main content
eLife logoLink to eLife
. 2019 Jun 27;8:e43578. doi: 10.7554/eLife.43578

Comment on 'AIRE-deficient patients harbor unique high-affinity disease-ameliorating autoantibodies'

Nils Landegren 1,2,, Lindsey B Rosen 3, Eva Freyhult 4,5, Daniel Eriksson 1,6, Tove Fall 7, Gustav Smith 8,9,10, Elise M N Ferre 3, Petter Brodin 8,11,12, Donald Sharon 13, Michael Snyder 10,13, Michail Lionakis 3, Mark Anderson 14, Olle Kämpe 1,6,15
Editors: Peter Rodgers16, Detlef Weigel17
PMCID: PMC6597240  PMID: 31244471

Abstract

The AIRE gene plays a key role in the development of central immune tolerance by promoting thymic presentation of tissue-specific molecules. Patients with AIRE-deficiency develop multiple autoimmune manifestations and display autoantibodies against the affected tissues. In 2016 it was reported that: i) the spectrum of autoantibodies in patients with AIRE-deficiency is much broader than previously appreciated; ii) neutralizing autoantibodies to type I interferons (IFNs) could provide protection against type 1 diabetes in these patients (Meyer et al., 2016). We attempted to replicate these new findings using a similar experimental approach in an independent patient cohort, and found no evidence for either conclusion.

Research organism: Human

Introduction

The immune system must be capable of mounting responses against foreign antigens of all possible sorts while at the same time remaining inert against self-components. Studies of a rare monogenic disorder – autoimmune polyendocrine syndrome type 1 (APS1) or autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy (APECED) and the mutated gene underlying the disease, Autoimmune regulator (AIRE) – have shed light on how reaction against self is avoided. The AIRE gene encodes a transcription factor that promotes thymic presentation of genes with limited peripheral tissue distribution (Anderson et al., 2002). AIRE-mediated antigen exposes naïve T-cells to self-molecules that otherwise would not be present in the thymus, and is therefore critical for the negative selection of autoreactive T-cells. Patients with APS1 develop multiple tissue-destructive manifestations and harbor autoantibodies targeting the affected tissues (Ahonen et al., 1990; Ferre et al., 2016; Söderbergh et al., 2004). A number of bona fide autoantigens have been identified in APS1 over the years, many of which are now used routinely in the clinical assessment of patients with APS1 and also in other autoimmune diseases (Alimohammadi et al., 2008; Alimohammadi et al., 2009; Ekwall et al., 1998; Kisand et al., 2010; Landegren et al., 2016a; Landegren et al., 2015; Meager et al., 2006; Puel et al., 2010; Shum et al., 2013; Winqvist et al., 1993; Winqvist et al., 1992). To gain a more comprehensive view on the autoantigen spectrum in APS1 we used a panel of 9000 full-length human proteins to perform a broad-scale mapping of autoantibody targets (Landegren et al., 2016b; Landegren et al., 2015). In the proteome array screen we replicated previously reported autoantigens and further identified and validated three novel major autoantigens. We limited our analyses to autoantibody targets that were shared between multiple patients and thereby showed statistically robust association with the patient group. We did not address autoantibody signals against targets that were observed only in one or a few patients, as we could not confidently separate these types of signals from the background of normal variation and technical noise. Indeed, elevated signals of this kind were observed both among patients and among the healthy controls.

Meyer et al. recently reported results from a screening experiment using a similar approach in APS1 patients, where the same protein panel was used to study autoantibodies in a comparable cohort of patients with APS1 and healthy controls (Meyer et al., 2016). We were surprised that their description of the autoantigen spectrum was very different from ours. Meyer et al. identified over 3700 proteins as autoantigens in APS1, corresponding to around 40% of the investigated panel. The healthy controls were found to largely lack high-level serum reactivity, which was used as evidence for the relevance of the autoantibody targets identified in the APS1 patients.

As the methods reported in the experiments by Meyer et al. were very similar to those we used, we asked whether differences in the approach to data analysis could explain the different conclusions. We noted that Meyer et al. applied a cutoff definition that could be expected to result in a skewed detection of autoantibody signals between patients and controls. To address this question, we used proteome array data for 51 APS1 patients and 21 healthy controls (Landegren et al., 2016b; Landegren et al., 2015) to reevaluate the autoantigen spectrum in APS1. In permuted datasets we found that the approach to data analysis used by Meyer et al. was strongly biased towards overestimating the number of autoantibody signals in the patient group. When we applied alternative statistical analyses without this inherent skewing effect, we found no support for the conclusion of broad antigen spectrum in APS1. Meyer et al. (2016) also proposed that subjects with APS1 that harbor neutralizing autoantibodies against type I interferons are protected against developing type 1 diabetes, one of the autoimmune manifestations of the disorder. We attempted to validate this finding in a larger cohort of APS1 subjects with type 1 diabetes and could find no evidence to support this proposed association.

Results

No evidence for widespread autoantibody reactivity in APS1 patients

We used our previously collected proteome array dataset to reevaluate the autoantigen spectrum in APS1. The data were generated by screening a panel of 9000 full-length human proteins (ProtoArray) with sera from 51 patients with APS1 and 21 healthy blood donors (Landegren et al., 2016b; Landegren et al., 2015). We followed the analytical approach used as closely as possible from the information available in the published article (Meyer et al., 2016). The fluorescence signal intensities were log-transformed and normalized using robust linear modeling based on the human-IgG and anti-human-IgG controls. Control spots and suspected printing contamination signals were sorted out (for further details, see Materials and methods). After filtering, 6540 unique proteins remained.

As a first step to ascertain if our dataset was comparable to the dataset in Meyer et al., we performed the same approach to statistical analysis as described to identify autoantibody signals in the APS1 patients and the healthy controls. Z-scores were calculated for all individuals over all proteins as the number of standard deviations above the mean value for the healthy controls. At low Z-score cutoffs (1, 2 and 3) the number of autoantibody signals was similar for patients and controls. However, at higher cutoffs the patient group deviated from the control group, and at cutoff level Z ≥ 5, autoantibody signals were only observed among the APS1 patients (Figure 1b). Our dataset thereby showed a similar result as the one by Meyer et al. when we used the same statistical analysis, specifically identifying high-level autoantibody signals among the APS1 patients. Investigation of established autoantigens in APS1 provided further support that the datasets in our study and the one by Meyer et al. were comparable (Figure 1—figure supplement 1).

Figure 1. The statistical analysis used by Meyer et al. is biased towards overestimation of the number of autoantibody signals in the patient group.

(A) We used our previously published proteome array dataset for 51 APS1 patients and 21 healthy controls to reevaluate the autoantigen spectrum in APS1. (B) To first determine if our dataset and the one by Meyer et al. were comparable, we applied the same statistical analysis as in Meyer et al. to identify autoantibody signals among the APS1 patients and healthy controls. Z-scores were calculated for all individuals over all proteins as the number of standard deviations above the mean value, where both mean and standard deviation were calculated based on the 21 healthy controls. Similar to the results by Meyer et al., high-level autoantibody signals (Z > 5) were specially observed among the patients. (C) To next determine whether the difference between patients and controls was true our could be explained by a biased statistical analysis we performed the same analysis in groups of subjects randomly assigned to represent ‘cases’ or ‘controls’. Fifty-one subjects were assigned to represent ‘cases’ and 21 subjects were assigned to represent ‘controls’, and autoantibody cutoffs were calculated based on the ‘controls’. The process was repeated 100 times. High-level autoantibody signals were specifically observed among the subjects assigned to represent ‘cases’. Mean values over the 100 permutations are presented in the table. (D) To determine whether the bias of the statistical analysis was dominating the overall results we applied a reversed cutoff, based on the values obtained for the APS1 patients instead of the healthy controls, to the true groups of patients and controls. The controls now showed high-level autoantibodies against greater number of proteins than the patients. In all panels, the average number of hits per individual for patients and healthy controls and the number of distinct proteins targeted in each group are shown for various Z-score cutoffs. The complexity factor was calculated as the number of distinct proteins divided by number of individuals divided by the average number of proteins per individual. NaN – not a number.

Figure 1.

Figure 1—figure supplement 1. Autoantibody signals for interferons in the protein array.

Figure 1—figure supplement 1.

Z-scores were calculated based on the mean and standard deviation of the healthy controls. p=APS1 patients (n = 51), C = controls (n = 21).
Figure 1—figure supplement 2. Simulation where an elevated outlier value at saturation level (65 000 signal units) was introduced in either the APS1 patient group or the control group for the array protein serum albumin.

Figure 1—figure supplement 2.

Z-scores were calculated based on the mean and standard deviation of the healthy controls. p=APS1 patients, C = controls.
Figure 1—figure supplement 3. Simulation where increasingly strong outliers (z-score 2, 3, …, 10) were introduced in random normal data for in total 51 ‘cases’ and 21 ‘controls’ respectively, illustrating how the cutoff is affected by elevated outliers in the control group but not by outliers in the case group.

Figure 1—figure supplement 3.

Horizontal bars indicate 3, 4, and five standard deviations above the ‘control’ group mean.
Figure 1—figure supplement 4. Analyses in permuted data from healthy controls reveal skewing effects in the analysis used by Meyer et al.

Figure 1—figure supplement 4.

To investigate whether there was an inherent skewing effect associated with the statistical method used by Meyer et al., analyses were performed using healthy blood donors randomly assigned to represent ‘cases’ or ‘controls’. Out of 21 healthy blood donors, eleven were assigned to represent ‘cases’ and ten were assigned to represent ‘controls’. Autoantibody cutoffs were calculated based on the mean and average for the blood donors assigned to represent ‘controls’. The process was repeated 100 times. High-level autoantibody signals were specifically detected among the blood donors that were assigned to represent ‘cases’. The average number of hits per individual for patients and healthy controls and the number of distinct proteins targeted in each group is shown for various cutoffs. The complexity factor was calculated as the number of distinct proteins divided by the number of individuals divided by the average number of proteins per individual.
Figure 1—figure supplement 5. Random selection of 10 protein identified as autoantigens in APS1 using the criteria applied by Meyer et al.

Figure 1—figure supplement 5.

(Z ≥ 3). Z-scores were calculated based on the mean and standard deviation of the healthy controls. p=APS1 patients (n = 51), C = controls (n = 21).
Figure 1—figure supplement 6. Targets identified using the analysis by Meyer et al. are low in signal and do not show enrichment for tissue-specific genes.

Figure 1—figure supplement 6.

To further evaluate the relevance of the targets identified using the statistical analysis by Meyer et al., we compared them against the established autoantigens in APS1 with respect to the number of autoantibody positive individuals, signal intensity, and enrichment for tissue-specific genes. The great majority of the new targets (Z ≥ 3) showed elevated level in only one or a few patients, with 37% of the targets representing single outlier values and only 26% being shared by more than three patients. The novel hits were further generally low in signal intensity compared to the established APS1 autoantigens. Looking at the top signal in the cohort for each target, the previously established autoantigens showed an average signal intensity of over thirty-seven thousand arbitrary fluorescence units while the new targets averaged at just below one thousand (A). We next investigated whether the new targets showed enrichment for tissue-specific genes. It has been shown that the AIRE gene promotes thymic expression of proteins with limited tissue distribution (Anderson et al., 2002). Consistent with the function of AIRE, the previously identified autoantigens in APS1 typically show tissue-specific expression. Therefore, studying enrichment of the tissue-specific genes could provide a way to determine if the new targets differed from a random sample of the array content. RNAseq-based gene expression data over multiple human tissues from The Human Protein Atlas (www.proteinatlas.com) (Uhlén et al., 2015) was used to study enrichment for gene expression category among the new targets as compared to the known autoantigen in APS1. The known autoantigens showed enrichment for the tissue-enriched (p<0.001), group-enriched (p<0.01) and not detected (p<0.001) gene categories, and none of them belonged to the mixed or expressed in all gene categories. The new targets, however, did not show enrichment for any category of tissue expression (B). We thus did not find any indication that the targets identified using the methods by Meyer et al. differed from a random sample of the studied protein panel.
Figure 1—figure supplement 7. No support for widespread autoantigen spectrum in APS1.

Figure 1—figure supplement 7.

As the statistical analysis used by Meyer et al. had proved to be biased in the comparison between patients and controls, the question remained whether or not the APS1 patients showed autoantibody responses against higher numbers of self-proteins than healthy subjects. We first looked at the distribution of autoantibody signals among patients and controls by applying different fixed signal cutoffs. The number of elevated IgG signals was higher in the APS1 patients than in the healthy controls, although the difference was small compared to that found by Meyer et al. (A). For example, at an arbitrary cutoff of 10,000 arbitrary fluorescence intensity units the APS1 patients displayed autoantibody responses against an average of 22 unique proteins as compared to 15 proteins in the healthy controls. Furthermore, the difference was almost completely eliminated when we excluded already known autoantigens – decreasing the number of autoantibody signals to an average of 16 proteins in the APS1 patients compared to the unchanged average of 15 proteins in the healthy controls (B). The results were similar for other cutoff values. We also compared the number of autoantibody signals between patients and controls by applying Z-score cutoffs. To allow for an unbiased detection of autoantibody signals, equally prone to register signals among patients and controls, we did not base the Z-score on the healthy controls as in the study by Meyer et al. but instead on the whole cohort. We further excluded the 10% highest and 10% lowest signals before calculating the cutoff, so that the cutoff level would not be affected by outlier values. This type of analysis is expected to detect elevated signals that occur in one or a few subjects, however, autoantibody signals that are present in many subjects are overlooked as the cutoff is consequently increased. Again we observed a small difference in the number of autoantibody signals between the patients and controls (C), which was further reduced by excluding the known autoantigens (D). Thereby, in contrast with results by Meyer et al. specifically identifying high-level autoantibody signals among the APS1 patients, we found only a small difference in the number of autoantibody signals between patients and controls. We also calculated the statistical significance that the number of autoantibody signals in APS1 differed from permutated data by comparing the number of distinct proteins with at least one hit (Z ≥ 3) among the APS1 cases to what is expected by random (when 21 samples are randomly assigned to represent controls and the remaining 51 represent cases) (E). The p-value was computed as the fraction of the permutations that result in as many distinct proteins as observed or more. At higher z-scores the numbers were borderline significant, showing a trend towards that the APS1 patients display higher numbers of autoantibody signals than the permutated data ‘cases’. The results were in line with the investigations in A-D. In all panels, the complexity factor was calculated as the number of distinct proteins divided by the number of individuals divided by the average number of proteins per individual. In panel A and B: Pat = APS1 patients, BG = healthy controls. In panel C and D: p=APS1 patients, C = healthy controls.

As a next step, we asked whether the difference in number of autoantibody signals between patients and controls was true or could be explained by a skewed data analysis. Meyer et al. used a cutoff based on the mean and standard deviation of the values obtained for the healthy controls to identify autoantibody signals. Consequently, elevated signals that occur among the controls increase the cutoff while elevated signals among the patients do not affect the cutoff. It could therefore be expected that elevated signals occurring sporadically in both patients and controls were detected unequally between the groups. To determine if this was the case, we first performed a simulation where we introduced an elevated outlier value in either the APS1 patient group or the control group for one of the proteins in the array. For this example, we chose the protein serum albumin, which did not show elevated signal for any patient or control. When we replaced one of the patient values with a signal value at saturation level (65,000 signal units), the cutoff was not affected and the elevated outlier scored as autoantibody positive at Z-score cutoffs of 3, 4 and 5 as expected (Figure 1—figure supplement 2). However, when we introduced an outlier at the same signal level in the control group the cutoff drastically increased, which gave the absurd result that the outlier scored as autoantibody negative at Z = 5. The example thereby revealed that even extremely elevated values in the control group could be missed in the analysis.

We also performed simulations where outliers were introduced in random normal data representing 51 ‘cases’ and 21 ‘controls’, which showed how outliers introduced in the controls were less likely to be detected (Figure 1—figure supplement 3). We next used our proteome array dataset to study groups of patients and controls randomly assigned to represent ‘cases’ or ‘controls’. Out of the 51 patients and 21 healthy blood donors in our cohort, we randomly assigned 51 subjects to represent ‘cases’ and 21 subjects to represent ‘controls’. The ‘control’ group was used to calculate the autoantibody cutoff levels. The process was repeated 100 times. The permuted datasets revealed a strong bias of the analytical method, specifically identifying high-level autoantibodies (Z ≥ 5) among the subjects assigned to represent ‘cases’ (Figure 1c).

We also performed a similar analysis limited to the healthy blood donors only, assigning eleven blood donors to represent ‘cases’ and the remaining ten to represent ‘controls’. Autoantibody signals were detected at high level in the blood donors that were assigned to represent ‘cases’ while those assigned to represent ‘controls’ did not show autoantibody signals above Z ≥ 3 (Figure 1—figure supplement 4). To better determine how the analytical bias influenced the overall results, we applied a reversed cutoff to our dataset based on the values obtained for the 51 APS1 patients instead of those for the 21 healthy controls. The healthy controls now turned out to show greater numbers of high-level (Z ≥ 5) autoantibody signals than the APS1 patients, revealing that the skewing effect of the data analysis was dominating the results (Figure 1d).

We further found that the targets identified using the data analysis by Meyer et al. were mostly low in signal, and in contrast to the established autoantigen, did not differ from a random sample of the protein panel when looking at the number of tissue-specific genes (Figure 1—figure supplements 56). When we applied alternative statistical analyses that treated cases and controls in a neutral way to compare the total number of autoantibody signals between the groups, we found only a minor difference between APS1 patients and controls (Figure 1—figure supplement 7). The difference between the groups was furthermore mostly explained by the previously identified APS1 autoantigens. Our investigations collectively suggested that the approach to data analysis used by Meyer et al. led to erroneous conclusions regarding the breadth of the antigen spectrum in APS1.

No association between neutralizing autoantibodies to interferons and type 1 diabetes in APS1

Autoantibodies against type 1 IFNs are present in almost all patients with APS1 and represent a valuable biomarker for the disease, even early in the course of the disease progression (Meager et al., 2006). Meyer et al. reported that type I IFN autoantibodies showed neutralizing effect in many but not all patients with APS1. The authors further found an inverse correlation between the capacity to neutralize IFNα and the development of type 1 diabetes in a small number of APS1 patients. Among 21 investigated APS1 patients, sera from all 13 patients without type 1 diabetes neutralized IFNα while sera from all eight patients with type 1 diabetes showed only low or negligible neutralization. No healthy controls or any other non-APS1 controls were included in the experiment. It was proposed that IFN autoantibodies could potentially protect patients with APS1 from developing type 1 diabetes.

We set out to investigate the proposed association between neutralizing type I IFN autoantibodies and type 1 diabetes. We investigated sera from three groups: 16 APS1 patients diagnosed with type 1 diabetes, 14 APS1 patients without type 1 diabetes from Sweden, Finland, the US and the UK, and five healthy controls. IFN neutralization was studied using a previously described cell-based assay (Gupta et al., 2016). Healthy control peripheral blood mononuclear cells were stimulated with IFNα, IFNω or IFNγ in the presence of APS1 patient or healthy control serum. We used the same serum concentration (10%) as previously applied to quantify interferon neutralizing autoantibodies in Gupta et al. (2016). CD14+ monocytes were then stained for intranuclear phospho-STAT1, which is a marker of IFN-dependent intracellular response. As expected, APS1 sera specifically neutralized IFNα- and IFNω- but not IFNγ-dependent pSTAT1 while healthy control sera did not show neutralization in any of the IFN stimulation assays. As control, serum from a patient with IFNγ autoantibodies and disseminated mycobacterial infection neutralized IFNγ- but not IFNα- or IFNω-dependent pSTAT1, and serum from a patient with IFNα and IFNω autoantibodies and thymoma neutralized IFNα- and IFNω- but not IFNγ-dependent pSTAT1 (data not shown). In contrast to the results by Meyer et al., we found strong neutralization of IFNα- and IFNω-dependent cellular response in all tested APS1 sera, with no difference between patients with type 1 diabetes compared to those without diabetes (Figure 2). Our results therefore did not confirm the proposed association between type 1 diabetes and neutralizing type I IFN autoantibodies in APS1.

Figure 2. No association between neutralizing interferon autoantibodies and type 1 diabetes in APS1.

Figure 2.

(A) Representative FACS plots showing IFNα-induced (blue), IFNω-induced (green), or IFNγ-induced (red) STAT1 phosphorylation in normal PBMC (gating on CD14 +monocytes) in the presence of 10% serum from a healthy control, an APS1 patient without type 1 diabetes (DM), or an APS1 patient with type 1 DM. (B) A stimulation index (ratio of stimulated to unstimulated geometric mean fluorescence values for pSTAT1 channel) was calculated for each serum sample tested in response to IFNα, IFNω, and IFNγ. No statistically significant difference was found between the means of APS1 patients with (n = 16) or without type 1 DM (N = 14) for IFNα- or IFNω-induced pSTAT1 (adjusted p value 0.9999 and 0.9976, respectively). IFNγ-induced pSTAT1 was not significantly different between healthy controls, APS1 patients without DM, or APS1 patients with DM. An unmatched 1-way ANOVA was used, correcting for multiple comparisons with Tukey’s test. Ns – not significant.

Figure 2—source data 1. Neutralizing interferon autoantibodies in APS1 patients with and without type 1 diabetes.
DOI: 10.7554/eLife.43578.011

Discussion

The study of the autoimmune spectrum in APS1 continues to provide potential novel insights into the general properties of autoimmunity. In this regard, Meyer et al. proposed two new major insights: i) that the spectrum of autoantibodies in APS1 is much broader than previously appreciated; ii) that neutralizing autoantibodies to type 1 IFNs could provide protection against type 1 diabetes in APS1 patients. Using a similar experimental approach on an independent cohort of APS1 subjects we found no evidence to support either conclusion. The finding of a broad autoantigen spectrum in APS1 was based on a biased statistical analysis prone to overestimate the number of autoantibody signals in the patient group. Collectively, we find no evidence to support the previous conclusion that almost half of the proteome is targeted by autoantibodies in APS1 patients and instead find that the autoantibody repertoire is far more restricted.

Studies of APS1 and its mouse model have provided important insights to the molecular mechanisms underlying immunological self-tolerance. The Aire-controlled genes have been studied comprehensively in mouse thymic tissue using omics technologies (Anderson et al., 2002; Sansom et al., 2014; Venanzi et al., 2008). Comprehensive studies of autoimmune targets in AIRE-deficient subjects represent an equally important source of information for understanding AIRE’s role in immune tolerance. Protein array technology provides the opportunity for mapping autoantibody targets on a broad scale. However, an inherent challenge with these types of studies, where thousands of tests are performed simultaneously, is to identify the relevant signals from a background of normal variation and technical noise. Unbiased comparison against controls becomes critical. Meyer et al. (2016) used the healthy controls to define normal signal ranges and to identify autoantibody responses. This type of approach is in many situations appropriate. In this case, however, it became a pitfall when a cutoff definition that requires the assumption that the controls show normal signals was used to compare the number of elevated signals between patients and controls. Background signals were unevenly recorded between patients and controls, resulting in the identification of an autoantigen spectrum dominated by the type of stochastic low-level signals that are also present among healthy individuals.

In our previous study of the autoantigen repertoire in APS1 we limited our analyses to autoantibody targets that were shared between multiple patients and thereby could be reliably identified from their statistical association with the patient group (Landegren et al., 2016b). Here we used other approaches to also address rare autoantigens in APS1. These studies revealed a only minor difference in the total number of autoantibody signals between APS1 patients and controls, which is in stark contrast with the results by Meyer et al. The healthy control group was small in both our study and the one by Meyer et al. A larger numbers of healthy controls would have provided better means to identify and evaluate the relevance of rare autoantibody signals in APS1 patients.

Meyer et al. also proposed that type I IFN autoantibodies may have disease ameliorating activity and protect patients with APS1 from developing type 1 diabetes, based on the absence of IFN neutralizing autoantibodies in a small number of APS1 subjects with type 1 diabetes (n = 8). In our independent investigation on a larger number of APS1 subjects with type 1 diabetes (n = 16) we could not repeat the reported association. Instead we found strong neutralization of the type I IFN-dependent cellular response in all investigated APS1 sera, with no difference in sera obtained from APS1 patients with or without type 1 diabetes. It is important to consider different explanations why the association between interferon neutralization and type 1 diabetes was not replicated, including differences in methodology. We used an established method and the same serum concentration as previously applied for quantifying neutralizing effect of interferon autoantibodies (Gupta et al., 2016). The positive and negative controls in the experiments also verified a reliable detection of neutralizing activity. Our results raise the need for further investigations to establish any association between IFN neutralizing autoantibodies and development of diabetes before embarking on in-depth investigations on targeting type 1 IFNs for the treatment or prevention of type 1 diabetes.

Materials and methods

Key resources table.

Reagent type
(species) or resource
Designation Source or
reference
Identifiers Additional
information
Biological sample (Homo sapiens) Serum samples from patients with APS1 from Sweden, Norway and Finland, and blood donor controls from Sweden PMID: 26084804 and 26830021
Biological sample
(Homo sapiens)
Serum samples from patients with APS1 from USA PMID: 27588307
Biological sample
(Homo sapiens)
Blood donor controls from NIH Blood Bank
Antibody Fluorophore-conjugated goat polyclonal anti-human IgG Thermo Fisher Scientific Alexa Fluor 647 Goat Anti-Human IgG; Cat#A21445, Concentration: 1:2000
Antibody Fluorophore-conjugated mouse monoclonal anti-pSTAT1 BD Biosciences Alexa Fluor 647 mouse anti-human Stat1 (pY701); Cat# 612597 5 uL/test
Antibody Fluorophore-conjugated mouse monoclonal anti-CD14 BD Biosciences FITC mouse anti-human CD14; Cat# 555397 2 uL/test
Recombinant protein Recombinant human interferon-alpha 2 PBL Assay Science Cat # 11101–2 Concentration:
10 ng/mL
Recombinant protein Recombinant human interferon-alpha Peprotech Cat# 300–02J Concentration:
10 ng/mL
Recombinant protein Recombinant human interferon-gamma 1b ActImmune NDC 42238-111-01 Concentration:
400 U/mL
Commercial assay or kit Protein microarray Thermo Fisher Scientific ProtoArray v5.0 (PAH0525020)
Commercial assay or kit Blocking buffer kit used in the protein array screening Thermo Fisher Scientific Blocking Buffer Kit (PA055)
Software, algorithm Software used for protein microarray scanning, alignment and data acquisition GenePix Pro microarray (v6.1)
Software FlowJo FlowJo (v10.5.3)
Software Prism GraphPad Prism (v6.0)

Study subjects

The patient cohort that was studied using protein arrays included individuals with APS1 from Sweden, Finland and Norway and healthy blood donors from Sweden, as described previously (Landegren et al., 2016b; Landegren et al., 2015). In the IFN neutralization studies, APS1 patients were also included from a second cohort at NIH. The diagnosis of type 1 diabetes mellitus was based on typical biochemical findings, including glycosuria, elevated plasma glucose levels, decreased plasma levels of insulin and C-peptide, and elevated hemoglobin A1c. Serum autoantibodies against GAD65 were analyzed in all patients. The study was approved by ethics boards of Stockholm (dnr: 2016/2553-31/2) and the NIAID, NIH Clinical Center and NCI Institutional Review Board committees. All patients have provided written informed consent for participation. The study was conducted in accordance with the Helsinki declaration.

NIH APS1 cohort

Sera were obtained from 12 APS1 patients enrolled in an IRB-approved protocol at the National Institute of Allergy and Infectious Diseases in the United States, as previously described (Ferre et al., 2016). Five APS1 patients had type 1 diabetes (representing all patients with type 1 diabetes in the NIH cohort of ~70 APS1 patients), and the remaining 7 APS1 patients did not have type 1 diabetes and were included as randomly-selected controls for functional evaluation of anti-IFN autoantibodies in serum. The diagnosis of type 1 diabetes mellitus was based on the presence of glycosuria, elevated plasma glucose levels, decreased plasma levels of insulin and C-peptide, and elevated hemoglobin A1c (Ferre et al., 2016). Eight patients were female, and five were children under the age of 18; the mean age of the 12 patients was 23.7 years (range 10–66 years). The majority (n = 11) originated from the US and one was from the Isle of Mann in the UK. The most common AIRE mutant allele, c.967_979del13, was found in homozygosity in four patients and in heterozygosity in combination with c.769C > T (n = 3), c.522_523ins13 (n = 1), c.1249_1250insC (n = 1), and c.1616C > T (n = 1). One patient was homozygous for c.769C > T and one patient was compound heterozygous for c.769C > T and c.789_789delC (Ferre et al., 2016). The most common clinical manifestations in the 12 patients were enamel hypoplasia (91.7%), hypoparathyroidism (83.3%), adrenal insufficiency (83.3%), chronic mucocutaneous candidiasis (75%), hepatitis (66.7%), pneumonitis (58.3%), Sjogren’s-like syndrome (41.7%), type 1 diabetes (41.7%), hypogonadism (41.7%), and hypothyroidism (8.3%). Autoantibodies against GAD65 were found in all five patients with type 1 diabetes and in 3 of the seven patients (42.9%) without type 1 diabetes.

APS1 cohort from Finland and Sweden

Eleven patients with type 1 diabetes and seven patients without type 1 diabetes from Finland and Sweden were included to the study on neutralizing IFN autoantibodies. The diagnosis of type 1 diabetes mellitus was based on typical biochemical findings, including the presence of glycosuria, elevated plasma glucose levels, decreased plasma levels of insulin and C-peptide, and elevated hemoglobin A1c. Seven out of the 18 patients were female. The mean age of the patients was 25 years (range 9–49 years). The most common clinical manifestations in the patients were chronic mucocutaneous candidiasis (100%), adrenal insufficiency (89%), hypoparathyroidism (83%), malabsorption (39%), gonadal insufficiency (33%), pernicious anemia (17%), alopecia (17%), vitiligo (11%) and autoimmune hepatitis (11%). GAD65 autoantibodies were detected in 5 out of 11 (45%) patients with type 1 diabetes and in 2 of the seven patients (29%) without type 1 diabetes.

Protein array screening

Human protein arrays containing ~9000 full-length human proteins (ProtoArray v5.0, Thermo Fisher) were screened with serum samples from patients with APS1 and healthy controls. The arrays were probed with serum at a concentration of 1:2000, and otherwise followed Invitrogen’s protocol for Immune Response BioMarker Profiling. A GenePix 4000B microarray scanner and the GenePix Pro microarray (v6.1) software was used for scanning, alignment and data acquisition. The proteome array dataset has been described previously (Landegren et al., 2016b; Landegren et al., 2015). The protein array screening experiment was not repeated. Suspected printing contamination artifacts were identified and excluded as previously described (Landegren et al., 2016b). No outliers were excluded.

Data analysis

Fluorescence intensities were background corrected and the mean value was computed over duplicate spots. Print-order artifacts were removed as follows: all spots with a Spearman correlation >0.8 to a spot with higher intensity that was printed right before were excluded. Values were log10-transformed and robust linear modeling (RLM) was applied to normalize the data based on human IgG and anti-human IgG controls as described previously (Sboner et al., 2009). RLM was implemented in the rlm function in the R package MASS. When more than one spot on the array was associated with the same gene name the mean value over the spots was calculated, such that each gene name was only represented once. The Z-scores were computed as the number of standard deviations away from the mean, where both mean and standard deviation were computed based on a set of ‘controls’. The Z-scores were computed for each protein separately.

We also calculated a robust Z-score based on trimmed mean and standard deviations, where the most extreme 10% of the values at each end were excluded when computing the mean and standard deviation. Enrichment of targets in any of the Human Protein Atlas classifications was measured using a hypergeometric test. The expression annotation categories were defined according to the following: tissue enriched – at least five-fold higher mRNA levels in a particular tissue as compared to all other tissues), group enriched - at least five-fold higher mRNA levels in a group of 2–7 tissues and tissue enhanced - at least five-fold higher mRNA levels in a particular tissue as compared to average levels in all tissues. The following proteins in the ProtoArray panel were treated as known autoantigens in the analyses: CYP1A2, DDC, GAD1, GAD2, GIF, IFNA1, IFNA13, IFNA14, IFNA16, IFNA17, IFNA2, IFNA21, IFNA4, IFNA5, IFNA6, IFNA8, IFNW1, IL17A, IL22, KCNRG, MAGEB2, PDILT, TGM4, TPH1 and TSGA10.

Functional evaluation of interferon autoantibodies

Healthy control peripheral blood mononuclear cells (PBMC; 1 × 107/mL) were either left unstimulated or stimulated for 15 min at 37°C with IFNα (10 ng/mL), IFNω (10 ng/mL), or IFNγ (400 U/mL) in the presence of 10% patient or control serum, as previously described (Gupta et al., 2016). Monocytes were identified by CD14 surface staining (BD Biosciences) before being fixed and permeabilized for intranuclear phospho-STAT1 (Y701; BD Biosciences). Data were collected using an LSRFortessa (BD Biosciences), analyzed using FlowJo software, and graphed with Prism6 software (GraphPad).

Acknowledgements

We thank Cindy Wong for critical review of the manuscript. This work was supported in part by the Division of Intramural Research of the NIAID/NIH

Funding Statement

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Contributor Information

Nils Landegren, Email: nils.landegren@ki.se.

Peter Rodgers, eLife, United Kingdom.

Detlef Weigel, Max Planck Institute for Developmental Biology, Germany.

Funding Information

This paper was supported by the following grants:

  • Swedish Research Council Formas to Olle Kämpe.

  • Novo Nordisk Foundation to Olle Kämpe.

  • ALF to Olle Kämpe.

  • Leona M. and Harry B. Helmsley Charitable Trust to Mark Anderson.

  • Division of Intramural Research, National Institute of Allergy and Infectious Diseases to Michail Lionakis.

  • Swedish Research Council to Olle Kämpe.

  • Swedish Society for Medical Research to Nils Landegren.

  • The Swedish Association of Endocrinologists to Nils Landegren.

  • Swedish Research Council to Nils Landegren.

  • Knut and Alice Wallenberg Foundation to Olle Kämpe.

Additional information

Competing interests

No competing interests declared.

Serves as founder and consultant for Personalis, is a member of the scientific advisory board of GenapSys, and a consultant for Illumina.

Is a board member of Olink Bioscience.

Author contributions

Conceptualization, Formal analysis, Writing—original draft.

Formal analysis, Investigation, Writing—review and editing, Analysis of interferon neutralising autoantibodies.

Formal analysis, Writing—review and editing, Bioinformatic analysis of protein array data.

Formal analysis, Writing—review and editing, Bioinformatic analysis.

Writing—review and editing, Critical review of the manuscript.

Writing—review and editing, Critical review of the manuscript.

Project administration, Writing—review and editing, Clinical characterisation of APS1 patients.

Writing—review and editing, Critical review of the manuscript.

Investigation, Writing—review and editing.

Resources, Funding acquisition, Writing—review and editing.

Conceptualization, Supervision, Funding acquisition, Writing—review and editing.

Conceptualization, Supervision, Funding acquisition, Writing—review and editing.

Conceptualization, Supervision, Funding acquisition, Writing—review and editing.

Ethics

Human subjects: The study was approved by ethics boards of Stockholm (dnr: 2016/2553-31/2) and the NIAID, NIH Clinical Center and NCI Institutional Review Board committees. All patients have provided written informed consent for participation. The study was conducted in accordance with the Helsinki declaration.

Additional files

Transparent reporting form
DOI: 10.7554/eLife.43578.012

Data availability

Source data has been provided Figure 2. Protein array data is available at ArrayExpress with accession E-MTAB-8097.

The following dataset was generated:

Nils Landegren, Donald Sharon, Michael Snyder, Olle Kämpe. 2019. Protein microarray-based autoantibody screening in 51 APS1 patients and 21 healthy controls. ArrayExpress. E-MTAB-8097

References

  1. Ahonen P, Myllärniemi S, Sipilä I, Perheentupa J. Clinical variation of autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy (APECED) in a series of 68 patients. New England Journal of Medicine. 1990;322:1829–1836. doi: 10.1056/NEJM199006283222601. [DOI] [PubMed] [Google Scholar]
  2. Alimohammadi M, Björklund P, Hallgren A, Pöntynen N, Szinnai G, Shikama N, Keller MP, Ekwall O, Kinkel SA, Husebye ES, Gustafsson J, Rorsman F, Peltonen L, Betterle C, Perheentupa J, Akerström G, Westin G, Scott HS, Holländer GA, Kämpe O. Autoimmune polyendocrine syndrome type 1 and NALP5, a parathyroid autoantigen. New England Journal of Medicine. 2008;358:1018–1028. doi: 10.1056/NEJMoa0706487. [DOI] [PubMed] [Google Scholar]
  3. Alimohammadi M, Dubois N, Sköldberg F, Hallgren A, Tardivel I, Hedstrand H, Haavik J, Husebye ES, Gustafsson J, Rorsman F, Meloni A, Janson C, Vialettes B, Kajosaari M, Egner W, Sargur R, Pontén F, Amoura Z, Grimfeld A, De Luca F, Betterle C, Perheentupa J, Kämpe O, Carel JC. Pulmonary autoimmunity as a feature of autoimmune polyendocrine syndrome type 1 and identification of KCNRG as a bronchial autoantigen. PNAS. 2009;106:4396–4401. doi: 10.1073/pnas.0809986106. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Anderson MS, Venanzi ES, Klein L, Chen Z, Berzins SP, Turley SJ, von Boehmer H, Bronson R, Dierich A, Benoist C, Mathis D. Projection of an immunological self shadow within the thymus by the aire protein. Science. 2002;298:1395–1401. doi: 10.1126/science.1075958. [DOI] [PubMed] [Google Scholar]
  5. Ekwall O, Hedstrand H, Grimelius L, Haavik J, Perheentupa J, Gustafsson J, Husebye E, Kämpe O, Rorsman F. Identification of tryptophan hydroxylase as an intestinal autoantigen. The Lancet. 1998;352:279–283. doi: 10.1016/S0140-6736(97)11050-9. [DOI] [PubMed] [Google Scholar]
  6. Ferre EM, Rose SR, Rosenzweig SD, Burbelo PD, Romito KR, Niemela JE, Rosen LB, Break TJ, Gu W, Hunsberger S, Browne SK, Hsu AP, Rampertaap S, Swamydas M, Collar AL, Kong HH, Lee CR, Chascsa D, Simcox T, Pham A, Bondici A, Natarajan M, Monsale J, Kleiner DE, Quezado M, Alevizos I, Moutsopoulos NM, Yockey L, Frein C, Soldatos A, Calvo KR, Adjemian J, Similuk MN, Lang DM, Stone KD, Uzel G, Kopp JB, Bishop RJ, Holland SM, Olivier KN, Fleisher TA, Heller T, Winer KK, Lionakis MS. Redefined clinical features and diagnostic criteria in autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy. JCI Insight. 2016;1:e88782. doi: 10.1172/jci.insight.88782. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Gupta S, Tatouli IP, Rosen LB, Hasni S, Alevizos I, Manna ZG, Rivera J, Jiang C, Siegel RM, Holland SM, Moutsopoulos HM, Browne SK. Distinct functions of autoantibodies against interferon in systemic lupus erythematosus: a comprehensive analysis of anticytokine autoantibodies in common rheumatic diseases. Arthritis & Rheumatology. 2016;68:1677–1687. doi: 10.1002/art.39607. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Kisand K, Bøe Wolff AS, Podkrajsek KT, Tserel L, Link M, Kisand KV, Ersvaer E, Perheentupa J, Erichsen MM, Bratanic N, Meloni A, Cetani F, Perniola R, Ergun-Longmire B, Maclaren N, Krohn KJ, Pura M, Schalke B, Ströbel P, Leite MI, Battelino T, Husebye ES, Peterson P, Willcox N, Meager A. Chronic mucocutaneous candidiasis in APECED or thymoma patients correlates with autoimmunity to Th17-associated cytokines. The Journal of Experimental Medicine. 2010;207:299–308. doi: 10.1084/jem.20091669. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Landegren N, Sharon D, Shum AK, Khan IS, Fasano KJ, Hallgren Å, Kampf C, Freyhult E, Ardesjö-Lundgren B, Alimohammadi M, Rathsman S, Ludvigsson JF, Lundh D, Motrich R, Rivero V, Fong L, Giwercman A, Gustafsson J, Perheentupa J, Husebye ES, Anderson MS, Snyder M, Kämpe O. Transglutaminase 4 as a prostate autoantigen in male subfertility. Science Translational Medicine. 2015;7:292ra101. doi: 10.1126/scitranslmed.aaa9186. [DOI] [PubMed] [Google Scholar]
  10. Landegren N, Pourmousa Lindberg M, Skov J, Hallgren Å, Eriksson D, Lisberg Toft-Bertelsen T, MacAulay N, Hagforsen E, Räisänen-Sokolowski A, Saha H, Nilsson T, Nordmark G, Ohlsson S, Gustafsson J, Husebye ES, Larsson E, Anderson MS, Perheentupa J, Rorsman F, Fenton RA, Kämpe O. Autoantibodies targeting a collecting Duct-Specific water channel in tubulointerstitial nephritis. Journal of the American Society of Nephrology. 2016a;27:3220–3228. doi: 10.1681/ASN.2015101126. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Landegren N, Sharon D, Freyhult E, Hallgren Å, Eriksson D, Edqvist PH, Bensing S, Wahlberg J, Nelson LM, Gustafsson J, Husebye ES, Anderson MS, Snyder M, Kämpe O. Proteome-wide survey of the autoimmune target repertoire in autoimmune polyendocrine syndrome type 1. Scientific Reports. 2016b;6:20104. doi: 10.1038/srep20104. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Meager A, Visvalingam K, Peterson P, Möll K, Murumägi A, Krohn K, Eskelin P, Perheentupa J, Husebye E, Kadota Y, Willcox N. Anti-interferon autoantibodies in autoimmune polyendocrinopathy syndrome type 1. PLOS Medicine. 2006;3:e289. doi: 10.1371/journal.pmed.0030289. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Meyer S, Woodward M, Hertel C, Vlaicu P, Haque Y, Kärner J, Macagno A, Onuoha SC, Fishman D, Peterson H, Metsküla K, Uibo R, Jäntti K, Hokynar K, Wolff ASB, Krohn K, Ranki A, Peterson P, Kisand K, Hayday A, APECED patient collaborative AIRE-Deficient patients harbor unique High-Affinity Disease-Ameliorating autoantibodies. Cell. 2016;166:582–595. doi: 10.1016/j.cell.2016.06.024. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Puel A, Döffinger R, Natividad A, Chrabieh M, Barcenas-Morales G, Picard C, Cobat A, Ouachée-Chardin M, Toulon A, Bustamante J, Al-Muhsen S, Al-Owain M, Arkwright PD, Costigan C, McConnell V, Cant AJ, Abinun M, Polak M, Bougnères PF, Kumararatne D, Marodi L, Nahum A, Roifman C, Blanche S, Fischer A, Bodemer C, Abel L, Lilic D, Casanova JL. Autoantibodies against IL-17A, IL-17F, and IL-22 in patients with chronic mucocutaneous candidiasis and autoimmune polyendocrine syndrome type I. The Journal of Experimental Medicine. 2010;207:291–297. doi: 10.1084/jem.20091983. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Sansom SN, Shikama-Dorn N, Zhanybekova S, Nusspaumer G, Macaulay IC, Deadman ME, Heger A, Ponting CP, Holländer GA. Population and single-cell genomics reveal the aire dependency, relief from polycomb silencing, and distribution of self-antigen expression in thymic epithelia. Genome Research. 2014;24:1918–1931. doi: 10.1101/gr.171645.113. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Sboner A, Karpikov A, Chen G, Smith M, Mattoon D, Dawn M, Freeman-Cook L, Schweitzer B, Gerstein MB. Robust-linear-model normalization to reduce technical variability in functional protein microarrays. Journal of Proteome Research. 2009;8:5451–5464. doi: 10.1021/pr900412k. [DOI] [PubMed] [Google Scholar]
  17. Shum AK, Alimohammadi M, Tan CL, Cheng MH, Metzger TC, Law CS, Lwin W, Perheentupa J, Bour-Jordan H, Carel JC, Husebye ES, De Luca F, Janson C, Sargur R, Dubois N, Kajosaari M, Wolters PJ, Chapman HA, Kampe O, Anderson MS. BPIFB1 is a Lung-Specific autoantigen associated with interstitial lung disease. Science Translational Medicine. 2013;5:ra139. doi: 10.1126/scitranslmed.3006998. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Söderbergh A, Myhre AG, Ekwall O, Gebre-Medhin G, Hedstrand H, Landgren E, Miettinen A, Eskelin P, Halonen M, Tuomi T, Gustafsson J, Husebye ES, Perheentupa J, Gylling M, Manns MP, Rorsman F, Kämpe O, Nilsson T. Prevalence and clinical associations of 10 defined autoantibodies in autoimmune polyendocrine syndrome type I. The Journal of Clinical Endocrinology & Metabolism. 2004;89:557–562. doi: 10.1210/jc.2003-030279. [DOI] [PubMed] [Google Scholar]
  19. Uhlén M, Fagerberg L, Hallström BM, Lindskog C, Oksvold P, Mardinoglu A, Sivertsson Å, Kampf C, Sjöstedt E, Asplund A, Olsson I, Edlund K, Lundberg E, Navani S, Szigyarto CA, Odeberg J, Djureinovic D, Takanen JO, Hober S, Alm T, Edqvist PH, Berling H, Tegel H, Mulder J, Rockberg J, Nilsson P, Schwenk JM, Hamsten M, von Feilitzen K, Forsberg M, Persson L, Johansson F, Zwahlen M, von Heijne G, Nielsen J, Pontén F. Proteomics. Tissue-based map of the human proteome. Science. 2015;347:1260419. doi: 10.1126/science.1260419. [DOI] [PubMed] [Google Scholar]
  20. Venanzi ES, Melamed R, Mathis D, Benoist C. The variable immunological self: genetic variation and nongenetic noise in Aire-regulated transcription. PNAS. 2008;105:15860–15865. doi: 10.1073/pnas.0808070105. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Winqvist O, Karlsson FA, Kämpe O. 21-hydroxylase, a major autoantigen in idiopathic Addison's disease. The Lancet. 1992;339:1559–1562. doi: 10.1016/0140-6736(92)91829-W. [DOI] [PubMed] [Google Scholar]
  22. Winqvist O, Gustafsson J, Rorsman F, Karlsson FA, Kämpe O. Two different cytochrome P450 enzymes are the adrenal antigens in autoimmune polyendocrine syndrome type I and Addison's disease. Journal of Clinical Investigation. 1993;92:2377–2385. doi: 10.1172/JCI116843. [DOI] [PMC free article] [PubMed] [Google Scholar]

Decision letter

Editor: Peter Rodgers1
Reviewed by: Christopher C Goodnow2

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Comment on 'AIRE-deficient patients harbor unique high-affinity disease-ameliorating autoantibodies'" to eLife for consideration as Scientific Correspondence. Your article, and the response to it from Kisand, Hayday and colleagues, have been reviewed by two peer reviewers (who have opted to remain anonymous), and the evaluation has been overseen by a Deputy Editor (Detlef Weigel) and the eLife Features Editor (Peter Rodgers).

I am pleased to be able to tell you that we have decided to accept for publication both your article and the response to it. The substantive comments in this decision letter will also be published as part of your article (and likewise for the response): you will be able to check these comments when you check the proof of your article.

You also need to make the following changes to your article:

[...]

Note: The editor handling this manuscript asked the reviewers to answer a number of questions: these questions are shown below in italics; the comments from the reviewers are shown in Roman.

Reviewer #1:

1) Are the elements of the present submission significant enough to merit publication in eLife? Do you have any substantive concerns about the elements of the article that challenge the findings of the original publication? Are the other elements of challenge significant enough to merit publication in eLife? Do you have any substantive concerns about these other elements?

There are two points of contention raised by the Landegren article in relation to the Meyer et al., 2016 Cell paper. The first is whether the extent of autoantigens in APECED patients is truly as broad as claimed by the Meyer publication. This is well challenged by the article, raising a 'data interpretation' debate. Landegren goes to great lengths to prove how the same data can be analyzed to show a broad range of serum autoreactivity or a restricted set of autoantigens. In an age of big data, this is entertaining and enlightening. Their concern is 100% warranted. The problem that is raised is that if one uses a small set of controls that are relatively homogeneous to define Z score thresholds for thousands of analytes, the analysis will favor outliers in a larger patient group. This is obvious, but not well understood. One simply should not use a small number of controls to define thresholds in these sorts of assays. Or if one does, then there must be a validation with a larger set or different assays. The point is well demonstrated. Landegren could have taken this further and more generally by making the point that one should not use a small number of controls to define thresholds as they simply are NOT sufficiently representative of the range of signals in health - that would be helpful for readers. What is perhaps less well done is to allow room for the possibility that there may be a broad reactivity since neither the original study nor the challenging analysis actually has sufficient controls to appropriately do the Z score approach and the alternative Landegren Fishers test approach is conservative. In other words, neither can actually conclude their view on broad vs restricted. Landegren should be critical of their own study by stating that they also had few controls and, therefore, used a more conservative approach to define the threshold, but that this is likely to be biased in the opposite direction to that of Meyer.

The second point of contention is the relationship of type 1 IFN neutralizing autoantibodies to diabetes in the APECED patients. Here, the challenge may be warranted but less convincing. It is important to examine a second cohort since the numbers in the Meyer paper were small. However, as pointed out in the formal response, Landegren failed to sufficiently quantify neutralization. They can conclude that the APECED patients who develop diabetes in their cohort have type 1 IFN neutralizing autoantibodies, but they cannot comment on whether the titer is the same or different to the APECED patients without diabetes since they have reached assay saturation in both groups. This is a pity since a suitably performed validation experiment would be a valuable contribution.

Reviewer #2:

1) Are the elements of the present submission significant enough to merit publication in eLife? Do you have any substantive concerns about the elements of the article that challenge the findings of the original publication? Are the other elements of challenge significant enough to merit publication in eLife? Do you have any substantive concerns about these other elements?

i) Are the elements of the present submission significant enough to merit publication in eLife?

Yes. The two main novel findings in the Cell paper, which have generated wide interest, appear not to be correct. The third finding that anti-interferon antibodies, which were already known to be very prevalent in AIRE deficiency, are of very high affinity due to affinity maturation, is not challenged by Landegren.

ii) Do you have any substantive concerns about the elements of the article that challenges the findings of the original publication?

No, the analysis performed by Landegren provide compelling evidence that the original publication in Cell employed a flawed normalization of the analysis of antibody binding to protein microarrays. As a result, the main conclusion in the Cell paper about the much wider breakdown of tolerance in AIRE deficiency, with antibodies recognizing private specificities unique to one or two patients, is erroneous.

In rebuttal, the authors of the Cell paper claim that a sampling of the thousands of private specificities were validated. There was no evidence for this in the published study, and unconvincing preliminary evidence here.

iii) Are the other elements of challenge significant enough to merit publication in eLife?

Yes. The second main novel conclusion of the Cell paper was that neutralizing antibodies to interferon were absent from the sera of AIRE deficient patients with Type 1 diabetes compared to those without. Here, studying a larger cohort, Landegren et al find neutralizing antibodies in AIRE deficient patients with or without Type 1 Diabetes. It is possible this reflects differences in assay methodology, but both methods appear equally valid and unlikely to explain the different results.

In rebuttal, the original authors point out that the data in the Cell paper were from serum titrations, and that if they repeat the experiments with serum diluted 1/10 they also now find neutralizing antibodies in both groups. In the Cell paper, the mean IC50 of neutralizing antibodies in patients without type 1 diabetes was on the order of 1/100,000, at which point there was a difference in titre. However in the data in the rebuttal, the difference in inhibitory activity is only apparent at one dilution (1/50) and not at 1/250 or 1/10. Since the rebuttal now also shows neutralizing antibodies are present in Type 1 diabetes patients at relatively small differences in titre, and since serum is undiluted in vivo, this appears to warrant a revision to the original conclusions in Cell that "those with T1D showed only low or negligible neutralization."

It is nevertheless difficult to extrapolate serum neutralization from in vitro to in vivo, so it would be important for Landegren et al to test interferon alpha neutralization at different serum dilutions to determine if there is a difference in titre in their cohort.

iv) Do you have any substantive concerns about these other elements?

No.

eLife. 2019 Jun 27;8:e43578. doi: 10.7554/eLife.43578.018

Author response


We repeat the reviewers’ points here in italic, and include our replies point by point, as well as a description of the changes made, in Roman.

Reviewer #1:

1) Are the elements of the present submission significant enough to merit publication in eLife? Do you have any substantive concerns about the elements of the article that challenge the findings of the original publication? Are the other elements of challenge significant enough to merit publication in eLife? Do you have any substantive concerns about these other elements?

There are two points of contention raised by the Landegren article in relation to the Meyer 2016 Cell paper. The first is whether the extent of autoantigens in APECED patients is truly as broad as claimed by the Meyer publication. This is well challenged by the article, raising a 'data interpretation' debate. Landegren goes to great lengths to prove how the same data can be analyzed to show a broad range of serum autoreactivity or a restricted set of autoantigens. In an age of big data, this is entertaining and enlightening. Their concern is 100% warranted. The problem that is raised is that if one uses a small set of controls that are relatively homogeneous to define Z score thresholds for thousands of analytes, the analysis will favor outliers in a larger patient group. This is obvious, but not well understood. One simply should not use a small number of controls to define thresholds in these sorts of assays. Or if one does, then there must be a validation with a larger set or different assays. The point is well demonstrated. Landegren could have taken this further and more generally by making the point that one should not use a small number of controls to define thresholds as they simply are NOT sufficiently representative of the range of signals in health - that would be helpful for readers. What is perhaps less well done is to allow room for the possibility that there may be a broad reactivity since neither the original study nor the challenging analysis actually has sufficient controls to appropriately do the Z score approach and the alternative Landegren Fishers test approach is conservative. In other words, neither can actually conclude their view on broad vs restricted. Landegren should be critical of their own study by stating that they also had few controls and, therefore, used a more conservative approach to define the threshold, but that this is likely to be biased in the opposite direction to that of Meyer.

REPLY: We appreciate the Reviewer’s recognition of efforts to bring clarity to the conflicting reports regarding the autoantigen repertoire in AIRE-deficient patients. It is encouraging to read that the Reviewer finds our points well demonstrated and enlightening.

We agree with the Reviewer that the small number of controls is an important limitation of both the study by Meyer et al. and ours. A larger number of controls would indeed have provided better means to evaluate what is truly patient-specific and what can be explained by normal variation. This would be especially important for determining the relevance (if any) of the type of rare autoantibody signals that are observed both among patients and healthy controls.

Despite this limitation, we are able to conclude the following:

1) Only a very restricted set of proteins (<1% of the protein panel) are targeted by autoantibodies in multiple APS1 patients. These major autoantigens, which can be readily identified by their statistical association with the patient group, are described in our previous article (Landegren et al., 2016).

2) There is no support for the Cell paper claim of broad antigen spectrum in APS1. Our analyses of permutated data clearly shows that their observed difference in numbers of autoantibody signals between patients and controls can be explained by a bias in the data analysis.

3) The total number autoantibody signals (also including rare autoantigens) is only slightly higher in APS1 compared to healthy control, which is in stark contrast with the Cell paper claim. When we apply different approaches that treat patients and controls in a neutral way to count the total number of autoantibody signals (either using Z-scores based on the values for both patients and controls where the most extreme values at each end have been excluded, or simply using fixed signal intensity cutoffs) we only find small difference between the patient group and the control group. This difference was furthermore mostly explained by the previously described autoantigens (Figure 1-figure supplement 7).

Changes made to the manuscript:

We have now added the following to the Discussion, explaining limitations of our study of the autoantigen repertoire:

In our previous study of the autoantigen repertoire in APS1 we limited our analyses to autoantibody targets that were shared between multiple patients and thereby could be reliably identified from their statistical association with the patient group (Landegren et al., 2016b). Here we used other approaches to also address rare autoantigens in APS1. These studies revealed a only minor difference in the total number of autoantibody signals between APS1 patients and controls, which is in stark contrast with the results by Meyer et al. The healthy control group was small in both our study and the one by Meyer et al. A larger numbers of healthy controls would have provided better means to identify and evaluate the relevance of rare autoantibody signals in APS1 patients.

The second point of contention is the relationship of type 1 IFN neutralizing autoantibodies to diabetes in the APECED patients. Here, the challenge may be warranted but less convincing. It is important to examine a second cohort since the numbers in the Meyer paper were small. However, as pointed out in the formal response, Landegren failed to sufficiently quantify neutralization. They can conclude that the APECED patients who develop diabetes in their cohort have type 1 IFN neutralizing autoantibodies, but they cannot comment on whether the titer is the same or different to the APECED patients without diabetes since they have reached assay saturation in both groups. This is a pity since a suitably performed validation experiment would be a valuable contribution.

REPLY: We here sought to provide an independent validation of the strong claims in the Cell paper regarding the existence of disease ameliorating autoantibodies in APS1. Specifically, we sought to repeat the observation that “patients without T1D collectively neutralized all IFNa subtypes, whereas those with T1D showed only low or negligible neutralization”.

In our validation study, we applied an established method to investigate autoantibody-mediated interferon neutralization in a larger cohort of APS1 patients with and without T1 diabetes. We used the same serum concentration as applied in the article by Gupta et al. for quantifying neutralization effect of interferon autoantibodies. The performance of the assay was verified using positive control sera (from a patient with IFNγ autoantibodies and disseminated mycobacterial infection, and from a patient with IFNα and IFNω autoantibodies and thymoma). In contrast to the Cell paper we included healthy blood donor sera as negative controls in our experiment, which would allow us to determine whether or not the APS1 without T1 diabetes truly showed negligible neutralization as previously claimed.

Our validation experiment showed strong and specific neutralization of IFNα- and IFNω in all tested APS1 sera, with no difference between patients with type 1 diabetes compared to those without diabetes. The results thus did not support the reported association with T1 diabetes status. Furthermore, our results from comparing APS1 to healthy control sera were in direct conflict with the Cell paper claim of low or negligible interferon neutralization in APS1 patients with T1 diabetes.

We understand from reading the comments of Reviewer 2 that the authors of the Cell paper can now only detect a difference in neutralization between patients with and without diabetes at certain serum dilutions. Hopefully further independent replication studies will be performed in light of the results presented in our Comment and in the Response to our Comment.

Changes made to the manuscript:

We have added the following to the Results, to clarify our reasoning behind the design of our experiments:

“We used the same serum concentration (10%) as previously applied to quantify interferon neutralizing autoantibodies in the study by Gupta et al.”

We have also added the following to the Discussion regarding limitations of our investigation:

“It is important to consider different explanations why the association between interferon neutralization and type 1 diabetes was not replicated, including differences in methodology. We used an established method and the same serum concentration as previously applied for quantifying neutralizing effect of interferon autoantibodies (Gupta et al., 2016). The positive and negative controls in the experiments also verified a reliable detection of neutralizing activity.”

Reviewer #2:

1) Are the elements of the present submission significant enough to merit publication in eLife? Do you have any substantive concerns about the elements of the article that challenge the findings of the original publication? Are the other elements of challenge significant enough to merit publication in eLife? Do you have any substantive concerns about these other elements?

i) Are the elements of the present submission significant enough to merit publication in eLife?

Yes. The two main novel findings in the Cell paper, which have generated wide interest, appear not to be correct. The third finding that anti-interferon antibodies, which were already known to be very prevalent in AIRE deficiency, are of very high affinity due to affinity maturation, is not challenged by Landegren.

ii) Do you have any substantive concerns about the elements of the article that challenges the findings of the original publication?

No, the analysis performed by Landegren provide compelling evidence that the original publication in Cell employed a flawed normalization of the analysis of antibody binding to protein microarrays. As a result, the main conclusion in the Cell paper about the much wider breakdown of tolerance in AIRE deficiency, with antibodies recognizing private specificities unique to one or two patients, is erroneous. In rebuttal, the authors of the Cell paper claim that a sampling of the thousands of private specificities were validated. There was no evidence for this in the published study, and unconvincing preliminary evidence here.

REPLY: It is encouraging to read the strong recognition of our work. We share the Reviewer’s view on the autoantibody results presented in the Cell paper.

iii) Are the other elements of challenge significant enough to merit publication in eLife?

Yes. The second main novel conclusion of the Cell paper was that neutralizing antibodies to interferon were absent from the sera of AIRE deficient patients with Type 1 diabetes compared to those without. Here, studying a larger cohort, Landegren et al find neutralizing antibodies in AIRE deficient patients with or without Type 1 Diabetes. It is possible this reflects differences in assay methodology, but both methods appear equally valid and unlikely to explain the different results.

In rebuttal, the original authors point out that the data in the Cell paper were from serum titrations, and that if they repeat the experiments with serum diluted 1/10 they also now find neutralizing antibodies in both groups. In the Cell paper, the mean IC50 of neutralizing antibodies in patients without type 1 diabetes was on the order of 1/100,000, at which point there was a difference in titre. However in the data in the rebuttal, the difference in inhibitory activity is only apparent at one dilution (1/50) and not at 1/250 or 1/10. Since the rebuttal now also shows neutralizing antibodies are present in Type 1 diabetes patients at relatively small differences in titre, and since serum is undiluted in vivo, this appears to warrant a revision to the original conclusions in Cell that "those with T1D showed only low or negligible neutralization."

It is nevertheless difficult to extrapolate serum neutralization from in vitro to in vivo, so it would be important for Landegren et al to test interferon alpha neutralization at different serum dilutions to determine if there is a difference in titre in their cohort.

iv) Do you have any substantive concerns about these other elements?

No.

REPLY: We thank the Reviewer for the valuable comments.

In our validation experiments we used an established method and the same serum concentration as previously applied for measuring neutralizing activity of interferon autoantibodies. The performance of the assay was verified using positive and negative controls (which was lacking in the original study). Our investigations revealed strong and specific neutralization of IFNα- and IFNω in all tested APS1 sera, with no difference between patients with type 1 diabetes compared to those without diabetes. The results thus did not support the reported association with T1 diabetes status.

It is surprising that difference in neutralization would only be detected in a very narrow range of concentration. We hope that further independent replication studies will be performed in light of the results presented in our Comment and in the Response to our Comment.

The following has been added to the Results to explain our reasoning behind choice of experimental setup:

“We used the same serum concentration (10%) as previously applied to quantify interferon neutralizing autoantibodies in the study by Gupta et al.”

We have also added the following to the Discussion to emphasize limitations:

“It is important to consider different explanations why the association between interferon neutralization and type 1 diabetes was not replicated, including differences in methodology. We used an established method and the same serum concentration as previously applied for quantifying neutralizing effect of interferon autoantibodies (Gupta et al., 2016). The positive and negative controls in the experiments also verified a reliable detection of neutralizing activity.”

References:

Gupta, S., Tatouli, I.P., Rosen, L.B., Hasni, S., Alevizos, I., Manna, Z.G., Rivera, J., Jiang, C., Siegel, R.M., Holland, S.M., et al. (2016). Distinct Functions of Autoantibodies Against Interferon in Systemic Lupus Erythematosus: A Comprehensive Analysis of Anticytokine Autoantibodies in Common Rheumatic Diseases. Arthritis & Rheumatology 68, 1677-1687.

Landegren, N., Sharon, D., Freyhult, E., Hallgren, A., Eriksson, D., Edqvist, P.H., Bensing, S., Wahlberg, J., Nelson, L.M., Gustafsson, J., et al. (2016). Proteome-wide survey of the autoimmune target repertoire in autoimmune polyendocrine syndrome type 1. Scientific Reports 6, 20104.

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Data Citations

    1. Nils Landegren, Donald Sharon, Michael Snyder, Olle Kämpe. 2019. Protein microarray-based autoantibody screening in 51 APS1 patients and 21 healthy controls. ArrayExpress. E-MTAB-8097

    Supplementary Materials

    Figure 2—source data 1. Neutralizing interferon autoantibodies in APS1 patients with and without type 1 diabetes.
    DOI: 10.7554/eLife.43578.011
    Transparent reporting form
    DOI: 10.7554/eLife.43578.012

    Data Availability Statement

    Source data has been provided Figure 2. Protein array data is available at ArrayExpress with accession E-MTAB-8097.

    The following dataset was generated:

    Nils Landegren, Donald Sharon, Michael Snyder, Olle Kämpe. 2019. Protein microarray-based autoantibody screening in 51 APS1 patients and 21 healthy controls. ArrayExpress. E-MTAB-8097


    Articles from eLife are provided here courtesy of eLife Sciences Publications, Ltd

    RESOURCES